Learning for Control: An Inverse Optimization Approach

We present a learning method to learn the mapping from an input space action space, which is particularly suitable when optimal decision with respect certain unknown cost function. use inverse optimization approach retrieve function by introducing new loss and hypothesis class of mappings. A tractable convex reformulation problem also presented. The effective for input-action in continuous input-output constraints, typically control systems. can be effectively transformed Model Predictive Control (MPC) behaviour case study mimic MPC presented, rather computationally heavy strategy. Simulation experimental results show effectiveness proposed approach.